Recommendation method and apparatus, and electronic device

Abstract

A recommendation method is provided, where the method includes: determining a user-event graph UEG based on training data, where the UEG includes a user node, an item node, a context node, and an intent node, and the intent node is separately connected to the user node, the item node, and the context node; performing graph convolution on a user feature, an item feature, a context feature, and an intent feature based on the UEG, to obtain a target feature set, where the target feature set includes a target user feature, a target item feature, and a target context feature; and recommending a target item to a target user based on the target feature set and a context feature of the target user in a target environment, where the target item is an item that matches the context feature in the target environment.

What is claimed is: 
     
         1 . A recommendation method, wherein the method comprises:
 determining a user-event graph (UEG) based on training data, wherein the UEG comprises: a user node, an item node, a context node, and an intent node, the intent node is separately connected to the user node, the item node, and the context node, the user node is a node on which a user feature of a user is located, the item node is a node on which an item feature of an item selected by the user is located, the context node is a node on which a context feature that affects selection of the item by the user is located, and the intent node is a node on which an intent feature of an intent of selecting the item by the user is located;   performing graph convolution on the user feature, the item feature, the context feature, and the intent feature based on the UEG, to obtain a target feature set, wherein the target feature set comprises a target user feature, a target item feature, and a target context feature; and   recommending a target item to a target user based on the target feature set and a context feature of the target user in a target environment, wherein the target item is an item that matches the context feature in the target environment.   
     
     
         2 . The method according to  claim 1 , wherein the intent feature is obtained based on the user feature and the context feature. 
     
     
         3 . The method according to  claim 2 , wherein there are a plurality of context features, and the method further comprises:
 separately determining weight values of the context features based on the user feature and the context features; and   determining the intent feature based on the context features and the weight values of the context features.   
     
     
         4 . The method according to  claim 1 , wherein the performing graph convolution on the user feature, the item feature, the context feature, and the intent feature based on the UEG, to obtain a target user feature specifically comprises:
 determining, based on a first intent feature at an (h−1) th  layer in a convolutional graph and an item feature that is associated with the first intent feature and that is at the (h−1) th  layer, a first user feature that is associated with the first intent feature and that is at an h th  layer in the convolutional graph; and   when a quantity of convolution times in the convolutional graph reaches a preset quantity of times, determining the target user feature based on first user features obtained at layers in the convolutional graph.   
     
     
         5 . The method according to  claim 4 , wherein the method further comprises:
 when there are a plurality of item features that are associated with the first intent feature and that are at the (h−1) th  layer, for any item feature associated with the first intent feature, separately determining, based on the any item feature and the first intent feature at the (h−1) th  layer, a second user feature that is associated with the first intent feature and that is at the h th  layer, to obtain a plurality of second user features;   determining, based on the plurality of second user features, the first user feature that is associated with the first intent feature and that is at the h th  layer; and   when the quantity of convolution times in the convolutional graph reaches the preset quantity of times, determining the target user feature based on the first user features obtained at the layers in the convolutional graph.   
     
     
         6 . The method according to  claim 1 , wherein the performing graph convolution on the user feature, the item feature, the context feature, and the intent feature based on the UEG, to obtain a target item feature specifically comprises:
 determining, based on a user feature that is associated with a first item feature and that is at the (h−1) th  layer in the convolutional graph and an intent feature that is associated with the first item feature and that is at the (h−1) th  layer, a first item feature at the h th  layer in the convolutional graph; and   when the quantity of convolution times in the convolutional graph reaches the preset quantity of times, determining the target item feature based on first item features obtained at the layers in the convolutional graph.   
     
     
         7 . The method according to  claim 6 , wherein the method further comprises:
 when there are a plurality of user features that are associated with the first item feature and that are at the (h−1) th  layer, for any user feature associated with the first intent feature, separately determining, based on the any user feature and the intent feature that is associated with the first item feature and that is at the (h−1) th  layer, a second item feature at the h th  layer, to obtain a plurality of second item features;   determining, based on the plurality of second item features, the first item feature at the h th  layer; and   when the quantity of convolution times in the convolutional graph reaches the preset quantity of times, determining the target item feature based on the first item features obtained at the layers in the convolutional graph.   
     
     
         8 . The method according to  claim 1 , wherein the performing graph convolution on the user feature, the item feature, the context feature, and the intent feature based on the UEG, to obtain a target context feature specifically comprises:
 determining, based on a user feature and an item feature that are associated with the first intent feature and that are at the (h−1) th  layer in the convolutional graph, a first intent feature at the h th  layer in the convolutional graph;   determining, based on the first intent feature at the h th  layer, a first context feature that is associated with the first intent feature and that is at the (h−1) th  layer, and a first user feature that is associated with the first intent feature and that is at the h th  layer, a first context feature at the h th  layer; and   when the quantity of convolution times in the convolutional graph reaches the preset quantity of times, determining the target context feature based on first context features obtained at the layers in the convolutional graph.   
     
     
         9 . The method according to  claim 8 , wherein the method further comprises:
 when there are a plurality of user features that are associated with the first context feature and that are at the h th  layer, for any user feature associated with the first context feature, separately determining, based on the any user feature, the first intent feature at the h th  layer, and the first context feature that is associated with the first intent feature and that is at the (h−1) th  layer, a second context feature at the h th  layer, to obtain a plurality of second context features;   determining, based on the plurality of second context features, the first context feature at the h th  layer; and   when the quantity of convolution times in the convolutional graph reaches the preset quantity of times, determining the target context feature based on the first context features obtained at the layers in the convolutional graph.   
     
     
         10 . The method according to  claim 1 , wherein the method further comprises:
 when one first context feature is associated with a plurality of first intent features, screening the plurality of first intent features in at least one convolution process, to obtain a required first intent feature, wherein a similarity value between the required first intent feature and the first context feature is greater than a preset similarity value.   
     
     
         11 . A recommendation apparatus, wherein the apparatus comprises:
 a graph construction module, configured to determine a user-event graph (UEG) based on training data, wherein the UEG comprises: a user node, an item node, a context node, and an intent node, the intent node is separately connected to the user node, the item node, and the context node, the user node is a node on which a user feature of a user is located, the item node is a node on which an item feature of an item selected by the user is located, the context node is a node on which a context feature that affects selection of the item by the user is located, and the intent node is a node on which an intent feature of an intent of selecting the item by the user is located;   a graph convolution module, configured to perform graph convolution on the user feature, the item feature, the context feature, and the intent feature based on the UEG, to obtain a target feature set, wherein the target feature set comprises a target user feature, a target item feature, and a target context feature; and   a recommendation module, configured to recommend a target item to a target user based on the target feature set and a context feature of the target user in a target environment, wherein the target item is an item that matches the context feature in the target environment.   
     
     
         12 . An electronic device, comprising:
 at least one memory, configured to store a program; and   at least one processor, configured to execute the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to perform:   determining a user-event graph (UEG) based on training data, wherein the UEG comprises: a user node, an item node, a context node, and an intent node, the intent node is separately connected to the user node, the item node, and the context node, the user node is a node on which a user feature of a user is located, the item node is a node on which an item feature of an item selected by the user is located, the context node is a node on which a context feature that affects selection of the item by the user is located, and the intent node is a node on which an intent feature of an intent of selecting the item by the user is located;   performing graph convolution on the user feature, the item feature, the context feature, and the intent feature based on the UEG, to obtain a target feature set, wherein the target feature set comprises a target user feature, a target item feature, and a target context feature; and   recommending a target item to a target user based on the target feature set and a context feature of the target user in a target environment, wherein the target item is an item that matches the context feature in the target environment.